The generation of reactive oxygen species (ROS) from an ultraviolet (UV)-exposed chemical can be an experimental indicator of phototoxic potential. The aim of the present study was to develop a new ROS assay using serum albumin to provide photosafety assessment under experimental biomimetic conditions. To assess assay robustness, a validation study on an albuminous ROS (aROS) assay was conducted with a focus on intra- and inter-day precisions and Z’-factor reflecting both the assay signal-to-noise ratio and variation associated with signal measurements. In the aROS assay on quinine HCl (200 μM), a typical phototoxic drug, both intra- and inter-day precisions (coefficient of variation; CV) were found to be below 4%, and the Z’-factors for singlet oxygen and superoxide suggested a large separation band between samples and blank signals. To evaluate the prediction capacity, the aROS and ROS assays were applied to 21 phototoxins and 10 non-phototoxic chemicals. Upon aROS assay on these model chemicals, the individual specificity was 100%, and the positive and negative predictivities were found to be 100% and 81.8%, respectively. The aROS assay can be employed for poorly soluble chemicals for which the ROS assay is unavailable. Comparing the ROS assay data, there seemed to be a photochemical transition of some chemicals in albuminous solution. A molecular interaction between albumin and chemical was also assessed by UV and fluorescent spectroscopic analyses, and the results suggested the limited relationship between the albumin–chemical interaction and the photochemical change. The aROS assay may allow photosafety assessment of new drug entities with a wide range of applicability partly under experimental biomimetic conditions. Copyright © 2013 John Wiley & Sons, Ltd.